Securing grid capacity and ordering backup generators may get an AI data centre project moving, but it doesn’t guarantee the resulting power system will remain resilient, adaptable and operational for the next 20 years.
By Steven Foley | 3 November 2026
Has the gigawatt become a vanity metric?
The data centre industry has a fascination with numbers – 100MW, 500MW, 1GW. The bigger the announced power requirement – the more impressive the project appears, but a gigawatt allocation is not an operating data centre, and neither is a generator purchase order.
The uncomfortable question for developers and operators is what happens between securing the power and delivering dependable power to the workload and that distinction is becoming increasingly important as AI changes the electrical characteristics of the data centre.
Higher-density computing creates larger concentrated loads and workloads can change rapidly. Cooling becomes more power-intensive, grid availability is becoming harder to secure and operators are being asked to deliver continuous availability while simultaneously managing cost, carbon and increasingly complex energy architectures. The old procurement model was simply not designed for all of this.
Is Power still being bought in pieces?
A conventional project can often involve several different organisations responsible for electrical design, switchgear, UPS systems, generators, controls, installation, commissioning and maintenance – whilst there is nothing inherently wrong with specialist suppliers – problems can arise where nobody ultimately owns the performance of the whole energy system.
A generator can work perfectly well while the controls connecting it to the rest of the facility fail. A UPS can perform exactly to specification while the wider power architecture is poorly coordinated, and a battery can provide flexibility but become operationally irrelevant if the site’s control system cannot use it intelligently.
In other words, individual components can all perform correctly while the system as a whole underperforms – and for an AI data centre, that’s not a procurement inconvenience, it’s an uptime risk.
Is the real product continuity?
This is where the market is beginning to move towards a different proposition: lifecycle accountability.
Rehlko is one example of this – following the rebranding of Kohler Energy, the company has positioned itself around integrated energy infrastructure spanning design, engineering, power generation, microgrids and lifecycle services. It says more than 90% of its deployed capacity is custom engineered for mission critical applications. The significance here is not necessarily the equipment itself – it is the attempt to take responsibility for the energy architecture beyond the point of delivery, and that distinction matters.
If the future data centre increasingly resembles an energy system with compute attached to it, operators will need suppliers who are capable of understanding how generation, storage, switching, controls, load management and maintenance interact over time. Should operators be asking a different question to whether they can supply the generator and start asking who is accountable when the entire power system doesn’t behave as expected?
AI makes the old model harder to defend
AI is accelerating this problem because the electrical infrastructure supporting dense compute cannot be treated as static. The facility may need to accommodate different generations of processors, changing rack densities, new cooling architectures and changing workload profiles which creates a lifecycle problem – the power system commissioned today could still be operating 15 or 20 years from now, but the IT equipment it was originally designed around almost certainly won’t be.
This is where microgrids, battery energy storage, intelligent controls and distributed generation become more than sustainability tools – they can become operational tools; Storage can provide flexibility, on site generation can provide additional resilience, controls can coordinate different energy sources and microgrids can provide greater autonomy when the grid becomes constrained.
None of those technologies are particularly valuable if they remain isolated assets so does the real value come from how intelligently the system operates as a whole?
Don’t confuse integration with buying everything from one supplier
There is an important distinction here; Lifecycle accountability does not necessarily mean that every component must come from one manufacturer and nor should operators surrender independent engineering judgement simply because a supplier offers an integrated package.
An operator can still use multiple specialist suppliers while demanding a clearly defined architecture, interfaces, performance obligations, commissioning standards and long-term accountability – that may ultimately be a more mature procurement model than simply attempting to minimise the number of purchase orders.
The operator takeaway
Operators should evaluate power partners against a much broader set of criteria – Can they:
- Design around the actual workload rather than a generic load profile?
- Integrate generation, UPS, BESS, controls and backup power?
- Support future changes in rack density?
- Manage the interaction between grid power and on-site generation?
- Provide meaningful monitoring and predictive maintenance?
- Guarantee clearly defined system-level performance?
- Remain accountable after commissioning?
- Provide a credible pathway towards lower-carbon operation without compromising resilience?
Those questions may become more important than the size of the equipment catalogue – and don’t confuse power capacity with power resilience. A gigawatt on a development plan, a yard full of generators and a collection of individually specified electrical components is not resilience – resilience comes from an energy architecture that can respond when conditions change, and from someone being accountable for making that architecture work.
The industry’s next procurement battleground might not be about who can manufacture the biggest generator or deliver the fastest equipment – it might be about who is prepared to own the outcome.
